Prediction and visualization of CYP2D6 genotype-based phenotype using clustering algorithms.

Area of Science:

• Pharmacogenomics
• Biotechnology
• Computational Biology

Background:

• Cytochrome P450 2D6 (CYP2D6) is crucial for drug metabolism.
• Predicting CYP2D6 phenotype from genotype is essential for personalized medicine.
• Existing methods may not fully capture the complexity of CYP2D6 genotype-phenotype relationships, especially for rare genotypes.

Purpose of the Study:

• To investigate the utility of unsupervised clustering algorithms for predicting CYP2D6 phenotypes from genotypes.
• To analyze dextromethorphan metabolic ratios (MRs) in a Korean population.
• To validate the predictive accuracy of clustering methods and color visualization for CYP2D6 activity.

Main Methods:

• Determined CYP2D6 genotypes and dextromethorphan MRs in 201 Koreans.
• Applied hierarchical and k-means clustering to a subset of 130 subjects.
• Utilized color visualization and receiver operating characteristic (ROC) curves for analysis and validation.

Main Results:

• Identified 23 distinct CYP2D6 genotypes, including rare ones.
• Clustering algorithms successfully classified CYP2D6 activity, aligning with known low (*5/*5) and high (duplicated alleles) activity groups.
• Validation in 71 subjects showed high concordance rates (92.8% for 3 clusters, 73.9% for 4 clusters).

Conclusions:

• Unsupervised clustering provides a valuable approach for predicting CYP2D6 phenotype from genotype.
• Color visualization offers a rapid method for interpreting genotype-phenotype relationships.
• This approach enhances understanding of CYP2D6 activity, including for rare genotypes, supporting personalized drug therapy.